Grasslands, comprising steppes, savannas, and prairies, cover a significant portion of Earth's land surface and contribute substantially to global primary production. While grasses (Poaceae) are the dominant biomass, other families, particularly daisies (Asteraceae), exhibit comparable or even greater diversity. The evolution of grasslands fundamentally altered terrestrial landscapes and spurred the adaptive radiation of numerous co-evolving organisms, including grazing mammals. The age of biomes is often estimated by examining the fossil record of representative taxa; for grasslands, this includes phytoliths and pollen data. However, phylogenetic analyses offer a more powerful approach, and this study leverages this approach using time-calibrated phylogenies of Poaceae and Asteraceae to estimate the timing of grassland expansion and explore the relationship between diversification and past environmental shifts. Analyzing hyper-diverse groups like Asteraceae (~23,000 species) and Poaceae (~11,000 species) presents challenges due to incomplete species sampling. The authors address this by developing a Bayesian approach that incorporates a more realistic model of species sampling, accounting for the non-uniform distribution of missing species in the phylogeny. This model uses the episodic birth-death process, allowing for shifts in speciation and extinction rates at different time points. This study focuses on testing correlations between diversification rates and two environmental variables: atmospheric CO2 concentration and average global paleo-temperature, using novel and existing environmentally-dependent diversification models. The goal is to refine our understanding of grassland origins within the context of past environmental changes.

Previous studies have estimated the age of grasslands using fossil records of grass phytoliths and pollen from grasses, daisies, and amaranths. Phylogenetic approaches, while powerful in studying other biomes like tropical rainforests, have been less applied to grasslands. Most previous studies of grassland evolution have primarily focused on the origins of C4 grasslands. This study, however, uses phylogenetic trees of Poaceae and Asteraceae to estimate the timing of grassland expansion. Existing approaches for inferring diversification rates can accommodate incomplete sampling, but they often use simplistic models of missing species distribution. The authors cite previous work demonstrating that biased species sampling can significantly affect diversification-rate estimates. They also reference existing environmentally-dependent diversification models that correlate diversification rates with environmental variables such as atmospheric CO2 and paleo-temperature, but these models often lack flexibility in handling incomplete data and complex diversification patterns.

The study utilized a large, time-calibrated phylogenetic tree for daisies (2723 sampled tips) and the largest currently available tree for grasses (3595 taxa). The daisy phylogeny was constructed using a combination of a backbone chronogram inferred from plastid DNA sequences and individual sub-trees representing the various subfamilies. The age of the nodes from backbone analysis were used to constrain the age of the eleven sub-trees. The grass phylogeny was taken from a previous study that inferred two chronograms with different calibration scenarios. For both phylogenies, a Bayesian approach was developed to estimate diversification rates through time. The approach combined the episodic birth-death process with environmentally-dependent diversification rates and an empirical taxon sampling model. The episodic birth-death process allows speciation and extinction rates to be constant within intervals but to shift instantly to new rates at rate-shift episodes. The empirical taxon sampling model uses taxonomic information about the distribution of unsampled species to improve the accuracy of rate estimates. Three prior models on diversification rates were compared: an uncorrelated lognormal (UCLN) prior, a Gaussian Markov random field (GMRF) prior, and a Horseshoe Markov random field (HSMRF) prior. These models allow for varying levels of autocorrelation in diversification rates. Environmental data on atmospheric CO2 concentration and average global paleo-temperature were obtained from published sources. The authors tested for correlations between diversification rates and these environmental variables using four environmentally-dependent diversification models: a fixed rate model, a model with UCLN variation, a model with GMRF variation, and a model with HSMRF variation. Model comparison was performed using Bayes factors. Finally, a simulation study was conducted to evaluate the robustness of the parameter estimates under the empirical taxon sampling model and to assess the bias introduced by assuming uniform taxon sampling. The simulations included trees generated under different diversification scenarios to test for false positives and evaluate the power of the method. The episodic birth-death process and environmentally-dependent diversification models were implemented in RevBayes, an open-source Bayesian phylogenetics software.

Analyses revealed a dramatic increase in diversification rates for both Asteraceae and Poaceae between the late Oligocene (~28 Mya) and early Miocene (~20 Mya). This shift was robust to various model assumptions, including the number of time intervals and different diversification rate prior models. The autocorrelated prior models (GMRF and HSMRF) were strongly favored compared to the uncorrelated model (UCLN). The diversification rate patterns were strongly influenced by the assumed incomplete taxon sampling, with the empirical taxon sampling model providing more accurate results compared to a uniform sampling assumption. The diversification rates for both families peaked between 20 Mya and 15 Mya, followed by a brief decrease before increasing again from the late Miocene (~10 Mya). A second analysis using a different calibration scenario for the grass phylogeny detected an earlier peak at 35–30 Mya. The timing of diversification rate shifts broadly agrees with those identified in previous studies for specific clades within Asteraceae and Poaceae, however, this study indicates that these were tree-wide, not lineage-specific events. Low diversification rates before ~35 Mya align with the scarcity of fossil records, while high rates after the Oligocene correspond with increased fossil diversity. The period of grassland diversification coincided with a significant decline in atmospheric CO2 during the Oligocene. Analyses of correlation between diversification rates and environmental variables showed a strong negative correlation between diversification rates and atmospheric CO2 for both families. The correlation with paleo-temperature was less clear, showing no significant support. The simulation study demonstrated the importance of using an accurate, empirical taxon sampling model, showing that incorrect assumptions about sampling lead to strongly biased diversification rate estimates. The study found a short-term decrease in diversification rates around 13–10 Mya, potentially linked to an increase in CO2 or the radiation of hypsodont grazers. The late Miocene increase in diversification is potentially due to the expansion of C4 grasses and hyper-diverse Asteraceae lineages.

The strong negative correlation between diversification rates of Asteraceae and Poaceae and atmospheric CO2 suggests a significant role for CO2 in driving grassland expansion and diversification. Low atmospheric CO2 can limit plant performance and lead to changes in vegetation structure. The study’s findings align with previous modeling work that links low CO2 with the expansion of grasslands at the expense of forests, particularly under conditions of increased aridity and decreased temperatures. While the CO2 hypothesis is central, other factors like decreasing temperatures, increasing aridity, seasonality, and grazing mammals likely played a role. The observed decrease in diversification rates during the mid-Miocene might be linked to the radiation of hypsodont grazers. The later increase in diversification rates might be attributed to the spread of C4 grasses and other hyper-diverse Asteraceae lineages. This study emphasizes the complex interplay between environmental and biological factors in shaping the evolution of grasslands.

This study provides strong evidence linking the rise of grasslands to a decline in atmospheric CO2 during the late Palaeogene. The development of a novel Bayesian phylogenetic approach, incorporating an empirical taxon sampling model and environmentally-dependent diversification rates, enabled a comprehensive analysis of large, incompletely sampled phylogenies. The negative correlation between diversification rates and CO2 levels suggests a significant influence of CO2 on grassland evolution. Further research could explore the relative contributions of other environmental and biological factors and could also focus on refining the phylogenetic trees used by incorporating more species and expanding the datasets for environmental variables. The findings presented highlight the importance of CO2 levels in shaping vegetation distributions and suggest that continued increases in atmospheric CO2 may have significant implications for future grassland diversity.

The study relied on existing phylogenies which, while large, are still incompletely sampled. While the empirical taxon sampling method improves the accuracy of diversification rate estimates, it does not fully eliminate the biases associated with incomplete sampling. Additionally, the analysis focuses on only two families (Poaceae and Asteraceae) and thus may not fully capture the overall dynamics of grassland diversification. The analysis uses smoothed curves for atmospheric CO2 and could potentially obscure fine-scale details in the environmental data. Future work could investigate potential alternative explanations for patterns observed.